Uneven balance of power between hypothalamicpeptidergic neurons in the control of feedingQiang Weia, David M. Krolewskia, Shannon Moorea, Vivek Kumara, Fei Lia, Brian Martina, Raju Tomerb,Geoffrey G. Murphya, Karl Deisserothc, Stanley J. Watson Jr.a, and Huda Akila,1

aMolecular and Behavioral Neuroscience Institute, University of Michigan, Ann Arbor, MI 48109; bDepartment of Biological Sciences, Columbia University,New York, NY 10027; and cDepartment of Bioengineering, Stanford University, Stanford, CA 94305

Contributed by Huda Akil, August 7, 2018 (sent for review February 6, 2018; reviewed by Olivier Civelli and Allen Stuart Levine)

Two classes of peptide-producing neurons in the arcuate nucleus(Arc) of the hypothalamus are known to exert opposing actions onfeeding: the anorexigenic neurons that express proopiomelano-cortin (POMC) and the orexigenic neurons that express agouti-related protein (AgRP) and neuropeptide Y (NPY). These neuronsare thought to arise from a common embryonic progenitor, butour anatomical and functional understanding of the interplay ofthese two peptidergic systems that contribute to the control offeeding remains incomplete. The present study uses a combinationof optogenetic stimulation with viral and transgenic approaches,coupled with neural activity mapping and brain transparencyvisualization to demonstrate the following: (i) selective activationof Arc POMC neurons inhibits food consumption rapidly in unsatedanimals; (ii ) activation of Arc neurons arising from POMC-expressing progenitors, including POMC and a subset of AgRPneurons, triggers robust feeding behavior, even in the face ofsatiety signals from POMC neurons; (iii) the opposing effects onfood intake are associated with distinct neuronal projection andactivation patterns of adult hypothalamic POMC neurons versusArc neurons derived from POMC-expressing lineages; and (iv) theincreased food intake following the activation of orexigenic neu-rons derived from POMC-expressing progenitors engages an ex-tensive neural network that involves the endogenous opioidsystem. Together, these findings shed further light on the dynamicbalance between two peptidergic systems in the moment-to-moment regulation of feeding behavior.

proopiomelanocortin | agouti-related protein | feeding |arcuate nucleus | satiety

Global epidemic obesity underscores the critical need forunderstanding how the brain regulates food intake (1).

Feeding is essential for survival and relies on mechanisms thatmonitor energy demands and metabolic status. But feeding be-havior is more complex, as it has a strong rewarding component(2) and can respond to environmental stimuli, including stressand social context (3). Therefore, it is critical to unravel themechanisms that modulate both the initiation and the cessationof feeding.It is well established that the hypothalamus plays an essential

role in the regulation of feeding behavior and contributes to theprocess of energy homeostasis (4, 5). Hypothalamic nuclei en-gaged in this regulation include the Arc, the paraventricularnucleus (PVN), the lateral hypothalamic (LH) area, the dorso-medial nucleus, and the ventromedial nucleus. Disruption of theventromedial hypothalamus induces hyperphagia and obesity,while disruption of the lateral hypothalamus causes hypophagiaand weight loss, suggesting the existence of ventromedial “satiety”and lateral “feeding” centers (6). Recent work has extended theseearly observations and has focused on the Arc of the hypothala-mus, a nucleus particularly important for the central regulation ofappetite, energy expenditure, and body weight (7, 8).Two classes of peptide-producing neuronal populations in the

Arc play a central and antagonistic role in the control of appetite—the orexigenic neurons that express AgRP and NPY promote food

consumption, while anorexigenic neurons that express POMC in-hibit feeding behavior. AgRP/NPY neurons are active when theenergy stores are not sufficient to meet systemic demands, therebyreleasing AgRP and promoting feeding (9, 10). POMC is a complexpolypeptide precursor cleaved into active peptides including mel-anocortins and β-endorphin (11). While β-endorphin is a long-acting opioid analgesic (12), the cosynthesized α-melanocyte stim-ulating hormone (α-MSH) blocks appetite (13). These two sets ofpeptidergic neurons in the Arc respond to nutrient and hormonalsignals from the periphery. Thus, leptin, a signal of positive energybalance, stimulates POMC neurons, leading to decreased foodintake while inhibiting the release of the orexigenic peptides fromneighboring AgRP/NPY neurons (10). Together, AgRP/NPY neu-rons and POMC neurons work coordinately to integrate signals ofenergy homeostasis and to regulate feeding behavior (8, 14).Evidence suggests that these two peptidergic systems that

control feeding emerge from a common cellular origin. A de-velopmental study has demonstrated that embryonic POMC-expressing progenitors not only give rise to adult POMC cells,but also differentiate into non-POMC neurons, including NPYneurons (15). This means transgenic studies that target POMCcells in early development can capture both adult populations,whereas adult delivery of viral vectors can target POMC cellsselectively. Consequently, we used optogenetic stimulation re-lying on both virally mediated and traditional transgenic ap-proaches to deliver blue light-activated channelrhodopsin 2(ChR2) to POMC neurons in the Arc. We coupled this stimulationwith behavioral assessments and anatomical analyses examining the

Significance

The interplay between the anorexigenic and orexigenic neuronsin the arcuate nucleus that contributes to the control of feedingremains elusive. Using optogenetic stimulation, we show thatactivation of POMC neurons rapidly inhibits feeding behavior infasted animals. However, simultaneous stimulation of bothPOMC neurons and a subset of the orexigenic neurons that ex-press AgRP is sufficient to reverse that inhibition and triggerintense feeding behavior. We used 3D imaging and functionalstudies to illuminate the anatomical underpinning of both theinhibitory and excitatory events. Our work suggests that trans-lational applications that aim to control appetite need to targetthe activation rather than the inhibition mechanisms.

Author contributions: Q.W., R.T., G.G.M., K.D., S.J.W., and H.A. designed research; Q.W.,D.M.K., S.M., V.K., F.L., and B.M. performed research; Q.W., D.M.K., S.M., V.K., and B.M.analyzed data; and Q.W. and H.A. wrote the paper.

Reviewers: O.C., University of California, Irvine; and A.S.L., University of Minnesota.

The authors declare no conflict of interest.

This open access article is distributed under Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0 (CC BY-NC-ND).1To whom correspondence should be addressed. Email: akil@umich.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1802237115/-/DCSupplemental.

Published online September 17, 2018.

www.pnas.org/cgi/doi/10.1073/pnas.1802237115 PNAS | vol. 115 | no. 40 | E9489–E9498

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 28

, 202

1

extent of activation and its consequences on cFos expressionthroughout the brain. Thus, we selectively activated neuronsderived from POMC-expressing progenitors in the Arc andascertained the net impact on food intake in adult mice. Wecontrasted the effect of activating only the POMC neurons ver-sus the impact of activating both POMC and AgRP/NPY neu-rons simultaneously on food consumption in mice. Our findingsindicate that selective activation of Arc POMC neurons inhibitsfood consumption rapidly in fasted animals whereas activation ofArc neurons arising from POMC-expressing progenitors resultsin a robust increase in food intake. We show that in the conflictbetween the drive to consume food and the rapid satiety controlthat inhibits feeding, the former dominates, resulting in an acuteincrease in feeding. These opposing effects are associated withdistinct projection and activation patterns. Thus, the regulationof feeding behavior involves a dynamic balance between the twopeptidergic systems.

ResultsCharacterization of Viral POMC-ChR2 Mice. To control hypothalamicPOMC neurons selectively, we used a Cre-dependent adeno-associated virus (AAV) vector carrying the gene encoding theblue light-activated cation channel ChR2 fused to enhancedyellow fluorescent protein (eYFP). This Cre-inducible double-floxed inverted ORF (DIO) AAV viral vector AAV5-EF1α-

DIO-ChR2(H134R)-eYFP (Fig. 1A) was microinjected into theArc of adult POMC-Cre transgenic mice (16). Upon trans-duction, Cre-expressing POMC neurons invert the ChR2-eYFPORF in irreversible means and ChR2-eYFP proteins are expressedunder control of the elongation factor 1α promoter (referred toas viral POMC-ChR2). This approach allowed us to selectivelystimulate Arc POMC neurons with blue light (473 nm). Wevalidated the cell-type selectivity of this viral approach in coronalArc sections of viral POMC-ChR2 mice. ChR2-eYFP specificallycolocalized with endogenous POMC neurons (Fig. 1B). On av-erage, 80% of POMC-positive neurons were positive for ChR2-eYFP in the virus injection sites (Fig. 1C). Of the ChR2-eYFPexpressing cells, 93% were colocalized by POMC immunostain-ing (Fig. 1C), demonstrating high specificity of ChR2 expressionin the POMC neurons of the Arc of the viral POMC-ChR2 mice.To confirm the functional impact of ChR2 activation in the viralPOMC-ChR2 mice, the Arc was illuminated with blue light whilethe animal explored an open field chamber. Expression of theneuronal activity marker, immediate early gene cFos, was evalu-ated in the Arc of the viral POMC-ChR2 mice with or withoutoptic stimulation. Activity in the region of virus injection was in-creased by approximately sevenfold following blue light stimula-tion [cFos-positive neurons per section: laser off (n = 3 mice),6.14 ± 1.27; laser on (n = 5 mice), 40.80 ± 3.74; unpaired two-tailed t test, t(6) = 6.83, P < 0.01]. Of cFos-positive cells, 77% were

POMC-CreEF-1a ChR2-eYFP

lox2272

loxP

ITR EF-1a WPRE polyA ITReYFP-ChR2A

ChR2-eYFP POMC MergeB

Labe

l 1 n

euro

ns a

lso

expr

essi

ng la

bel 2

(%)

(1) POMC(2) ChR2 (2) POMC

(1) ChR20

20

40

60

80

100C

ChR2-eYFP cFos MergeD E

Off On

% c

Fos+

in C

hR2-

expr

essi

ng ****

0

20

40

60

80

100

Fig. 1. Characterization of viral POMC-ChR2 mice. (A) Schematic of the Cre-dependent AAV. The ChR2-eYFP gene is doubly flanked by two sets of in-compatible lox sites. Upon delivery into POMC-Cre transgenic mouse line, ChR2-eYFP is inverted to enable transcription from the EF-1α promoter. (B) Rep-resentative images showing cell-specific ChR2-eYFP expression (green) in POMC neurons (red) in the Arc of the hypothalamus. (Scale bar, 50 μm.) (C) Statisticsof expression in POMC neurons (n = 6 mice, 894 POMC neurons, 738 ChR2-eYFP neurons). Eighty percent of POMC neurons expressed in ChR2-eYFP neurons;93% of ChR2-eYFP neurons expressed in POMC neurons. Mean ± SEM. (D) Blue light illumination of the Arc led to induction of cFos (red) in ChR2-eYFP-positive neurons (green). (Scale bar, 20 μm.) (E) Statistics of cFos-positive neurons in ChR2-eYFP expressing cells at implant region of the Arc with or withoutoptic stimulation. Six percent of cFos-positive neurons expressed in ChR2-eYFP neurons in the no-optic stimulation group (n = 3 mice, 55 cFos-positive neurons,410 ChR2-eYFP neurons) versus 77% in light-stimulation group (n = 5 mice, 538 cFos-positive neurons, 554 ChR2-eYFP neurons). Unpaired two-tailed t test:t(6) = 11.77, ****P < 0.0001. Mean ± SEM.

E9490 | www.pnas.org/cgi/doi/10.1073/pnas.1802237115 Wei et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 28

, 202

1

colabeled by ChR2-eYFP staining (Fig. 1 D and E). In contrast,6% of cFos-positive cells were colocalized in ChR2-eYFP stainingin the no-stimulation group (Fig. 1E). These results indicate se-lective and effective local activation of POMC neurons in theviral POMC-ChR2 mice.

Characterization of Transgenic POMC-ChR2 Mice. To target ChR2expression selectively in POMC progenitors, the POMC-Cremouse line was crossed to a conditional mouse line containingRosa26-CAG-stopflox-ChR2(H134R)-eYFP (17). Offspring ofthe double transgenic mouse line are hereafter referred to astransgenic POMC-ChR2 (Tg POMC-ChR2) (Fig. 2A). Adult TgPOMC-ChR2 mice exhibited robust expression of ChR2-eYFPin the Arc (Fig. 2B). We found that both POMC-positive (Fig. 2C)and AgRP-positive (Fig. 2D) neurons were detected in ChR2-eYFP expressing cells, indicating that the POMC-expressing line-age gives rise to two opposing neuronal populations in the Arcduring development. This is consistent with the view that embry-onic POMC-expressing progenitors differentiate into AgRP/NPYexpressing cells (15). Furthermore, our immunostaining studyrevealed that this group of AgRP-positive neurons derived fromPOMC-expressing progenitors represents a subset of AgRP neu-rons in the Arc (Fig. 2D).Ex vivo, whole-cell current-clamp recording revealed that

ChR2-eYFP-positive cells fired action potentials in response torepeated light-stimulation patterns used in the hot-plate andfood intake tests of the current study (Fig. 2 E and F), indicatingthat ChR2-expressing cells were capable of responding to lightpulses during the entire behavioral testing sessions and they werewell tolerant to light stimulations.Blue light stimulation of the Arc of Tg POMC-ChR2 mice in-

duced a significant increase in cFos expression in the region belowthe implanted optic fiber [cFos-positive neurons per section: laseroff (n = 3 mice), 4.06 ± 0.82; laser on (n = 3 mice), 49.17 ± 1.12;unpaired two-tailed t test, t(4) = 32.5, P < 0.0001]. Sixty-onepercent of ChR2-eYFP expressing neurons were activated in thelight-stimulation group compared with 5% in the group withoutstimulation (Fig. 2G). Using triple-label immunostaining, wefound that both POMC-positive and AgRP-positive neurons (Fig.2 H and I) were activated by light stimulation as assessed by in-duction of the cFos. These results demonstrate that, in adult TgPOMC-ChR2 mice, ChR2 proteins are expressed in POMC orAgRP neurons and these two opposing neuronal populations aresimultaneously activated by light stimulation.

Rapid Inhibition of Feeding Behavior in Fasted Viral POMC-ChR2 Mice.We first examined whether activating POMC neurons by opto-genetic stimulation was sufficient to reduce food intake in viralPOMC-ChR2 mice. Food consumption was measured for 30 mineach day for three consecutive days (day 1, blue laser off; day 2,blue laser on; day 3, blue laser off). Mice were food-deprived for4 h each day before the food intake test. The laser stimulationparameters used on day 2 were 15-ms light pulses at 10 Hz for30 s every minute for 30 min. Stimulation of Arc POMC neuronsin fasted viral POMC-ChR2 mice resulted in a significant re-duction in food intake (Fig. 3A). In addition, we studied theimpact on pain regulation since this is one of the most classicaland well-established effects of β-endorphin (18). The latency forthe mice to lick their paws was measured in the hot-plate test.Viral POMC-ChR2 mice showed an increased nociceptivethreshold following light stimulation (Fig. 3B). This analgesiceffect induced by light stimulation could be blocked by pre-treatment with the opioid antagonist naloxone (Fig. 3C), sug-gesting that β-endorphin was released by activation of ArcPOMC neurons. Together, these results demonstrate that func-tional neuropeptides in POMC neurons can be released by lightstimulation, and activation of Arc POMC neurons leads to acuteinhibition of food intake in fasted animals.

Rapid and Robust Increase in Food Intake in Tg POMC-ChR2 Mice. InTg POMC-ChR2 mice, ChR2 protein was expressed in neuronsderived from POMC-expressing progenitors in the Arc includingPOMC- and AgRP-positive neurons. We also demonstrated thatboth POMC and AgRP neurons are simultaneously activated bylight stimulation in the Arc of Tg POMC-ChR2 mice. Hot-plateand food intake tests were performed in this double transgenicmouse line. Activation of the Arc in Tg POMC-ChR2 mice bylight stimulation induced an acute analgesic effect in the hot-plate test (Fig. 4A), which could be blocked by pretreatment withnaloxone (Fig. 4B). This suggests that POMC neurons in the TgPOMC-ChR2 mice are functionally activated by optogeneticstimulation. Latency for single transgenic littermate control micewithout ChR2-eYFP expression to lick their paws was unchangedfollowing optogenetic stimulation (Fig. 4A). Surprisingly, TgPOMC-ChR2 mice displayed a vigorous increase in food intakein response to light stimulation in the Arc (Fig. 4C). There wasno change in food consumption in control mice following pho-tostimulation (Fig. 4C). These results demonstrate that activa-tion of the Arc in Tg POMC-ChR2 mice leads to a dramaticincrease in food intake, counteracting the effect of activation ofPOMC neurons in the same area. Thus, the orexigenic neuronsthat drive food consumption clearly override the anorexigenicneurons that inhibit feeding in this animal model.

Distinct Neuronal Projections Between Viral and Tg POMC-ChR2 MiceVisualized via 3D Imaging. A number of recently developed braintransparency techniques permit the 3D visualization of fluorescent-labeled molecules in genetically modified animals. The CLARITY(19–21) and iDISCO (22) methods are two such approachescapable of illuminating neuronal networks that may offer in-sight into brain connectivity in large millimeter-thick tissuesamples. To better understand the potential association be-tween behavior phenotypes and related brain circuitries, weimplemented the CLARITY and iDISCO methods for the viraland Tg POMC-ChR2 animal brain samples, respectively. Inthe viral POMC-ChR2 mice, anti-GFP antibody-labeled fibersoriginating from the Arc were observed throughout the extentof analyzed samples (Fig. 5 A and B and Movies S1 and S2).Our findings show that the Arc POMC neurons send efferentprojections to various regions, including: (i) ventral to themedian eminence; (ii) rostral to the medial preoptic nuclei, thebed nucleus of the stria terminalis, the nucleus accumbens(NAcc), the lateral and medial septum, and lateral habenularnucleus; (iii) dorsal to the thalamic periventricular nucleus; (iv)lateral to the amygdala; (v) caudal to the brainstem regions, in-cluding the periaqueductal gray; and (vi) ventrocaudal to themammillary nucleus (SI Appendix, Table S1). Arc POMC neuronsproject extensively to paraventricular, dorsomedial, and lateralhypothalamic nuclei. Additionally, anterior and periventricularhypothalamic areas and zona incerta also receive direct anddense inputs from the Arc. Nucleus of the vertical limb of thediagonal and the horizontal limb of the diagonal band also re-ceive direct projections from the Arc POMC neurons (SI Ap-pendix, Table S1).In contrast, in neurons derived from POMC-expressing pro-

genitors in the Arc of the Tg POMC-ChR2 mice, the majority oftheir projections remain within hypothalamic areas (Fig. 6 A andB and Movies S3 and S4). Here, anti-GFP antibody-labeled fi-bers were observed terminating in the paraventricular, periven-tricular, anterior, and the dorsomedial hypothalamus. It isremarkable that the most qualitatively robust innervation wasseen within the lateral hypothalamic nucleus (Fig. 6 A and B andMovies S3 and S4). We also observed the anti-GFP antibody-labeled signals in the dentate gyrus of the hippocampus (Fig. 6A)in Tg POMC-ChR2 mice. This is not surprising, given that ex-pression of POMC follows a dynamic pattern during develop-ment and POMC expression in some regions of an embryo is

Wei et al. PNAS | vol. 115 | no. 40 | E9491

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 28

, 202

1

XCAG WPREStopRosa26

ChR2-eYFP

loxP

POMC Cre

A

Arc

B

ChR2-eYFP POMC Merge

C

ChR2-eYFP AgRP Merge

D

POMCChR2-eYFP cFos MergeH

MergecFosAgRPChR2-eYFPI

G

% c

Fos+

in C

hR2-

expr

essi

ng

Off On0

20406080

100

***

E1

2 3

40 mV25 ms

1 2 3

Stimulation

40 mV30 s

1 2 34

40 mV25 ms

1 2 3 4

40 mV5 s

Stimulation (Minute 1) Stimulation (Minute 30)

F

Fig. 2. Characterization of Tg POMC-ChR2 mice. (A) Genetic design for ChR2-eYFP expression in the Arc. (B) Expression of ChR2-eYFP in the Arc. (C) Rep-resentative images showing colocalization of POMC-positive neurons with ChR2-eYFP expressing cells in the Arc. Inset shows ChR2-eYFP expression (green) inPOMC neurons (red). (Scale bar, 50 μm.) (Scale bar in Inset, 20 μm.) (D) Representative images showing colocalization of AgRP-positive neurons with ChR2-eYFP expressing cells in the Arc. Inset shows ChR2-eYFP expression (green) in a subgroup of AgRP neurons (red). Yellow arrows point to ChR2-eYFP-positiveAgRP neurons; white arrows point to ChR2-eYFP-negative AgRP neurons. (Scale bar, 50 μm.) (Scale bar in Inset, 20 μm.) (E and F) Representative whole-cellcurrent-clamp recordings of ChR2-eYFP-positive neurons in the Arc from acute slices. To replicate the stimulation pattern used in the hot-plate test, theneuron was stimulated with 15-ms light pulses at 10 Hz for 6 min (indicated by blue bar). After the 6-min period of action potential firing, the light was turnedoff to show the specificity of the response to light stimulation. Action potential firing resumed at 10 Hz when the light in turned back on. Insets show in-dividual action potential firing at 10 Hz at the times indicated by the numbers. To replicate the stimulation pattern used in the food intake study, the neuronwas stimulated with 15-ms light pulses at 10 Hz for 30 s (indicated by blue bar) every other 30 s for 30 min. For simplicity, only the first stimulation period(during minute 1; Left trace) and the last stimulation period (during minute 30; Right trace) are shown. Insets show individual action potential firing at 10 Hzat the times indicated by the numbers. n = 25 cells, 15 slices from 8 mice. (G) Statistics of cFos-positive neurons in ChR2-eYFP expressing cells in the regionbelow the implanted optic fiber in the Arc with or without optic stimulation. Five percent of cFos-positive neurons expressed in ChR2-eYFP neurons in the no-optic stimulation group (n = 3 mice, 30 cFos-positive neurons, 522 ChR2-eYFP neurons) versus 61% in the light-stimulation group (n = 3 mice, 543 cFos-positiveneurons, 901 ChR2-eYFP neurons). Unpaired two-tailed t test: t(4) = 13.63, ***P < 0.001. Mean ± SEM. (H) Representative images showing the induction ofcFos (white) in POMC-positive (red) ChR2-eYFP expressing neurons (green) (yellow arrows) following blue light stimulation in the Arc. POMC-positive neuronswere immunostained with anti–β-endorphin antibody. (Scale bar, 20 μm.) (I) Representative images showing the induction of cFos (white) in AgRP-positive(red) ChR2-expressing neurons (green) (yellow arrows) following light stimulation in the Arc. (Scale bar, 20 μm.)

E9492 | www.pnas.org/cgi/doi/10.1073/pnas.1802237115 Wei et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 28

, 202

1

transient (23), including the granule cells of the dentate gyrus(24). These results demonstrate distinct neuronal projectionpatterns in these two animal models.

Distinct Neuronal Activation Patterns Between Viral and Tg POMC-ChR2 Mice. Given the opposite effects on food intake betweenviral POMC-ChR2 and Tg POMC-ChR2 mice, we evaluated theneuronal activation as determined by cFos mRNA expressionfollowing light stimulation in the Arc in these two animal models.Activation of POMC neurons of the Arc in viral POMC-ChR2 mice led to an increase in cFos in limited brain regions(Fig. 5C), including cerebral cortex, lateral septum, PVN, dor-somedial hypothalamus, Arc, and the medial posterior part ofthe Arc (SI Appendix, Table S2). In contrast, activation of theArc in Tg POMC-ChR2 mice resulted in an intensive cFos re-sponse in broader areas of the brain (Fig. 6C), including cerebralcortex, prefrontal cortex (PFC), caudate-putamen (CPu), NAcc,septum, lateral thalamus, PVN, LH, dorsomedial hypothalamus,and Arc (SI Appendix, Table S3). These results suggest that this

acute increase in food intake following activation of the Arc inTg POMC-ChR2 mice engages an extensive neuronal system inthe brain.

Role of Endogenous Opioids in Increased Feeding in Tg POMC-ChR2Mice. We further investigated the role of endogenous opioids inincreased food intake observed in Tg POMC-ChR2 mice. Thisincreased feeding could be largely blocked by pretreatment withnaloxone in a dose-dependent manner (Fig. 7A), indicating thedriving force to consume food can be partially abolished byblocking opioid receptor signaling in this animal model. Fol-lowing naloxone pretreatment, increased cFos mRNA expressionin LH, dorsomedial hypothalamus, and the Arc was observed inresponse to light stimulation in the Arc of the Tg POMC-ChR2 mice (Fig. 7B and SI Appendix, Table S4). It should benoted that the induction of neuronal activity in the Arc and LHwas reduced by 63% and 52%, respectively, following naloxonepretreatment (Arc: optical density of cFos induction 0.74 versus0.27 relative to respective control; LH: optical density of cFos

Late

ncy

to li

ck p

aws

(sec

)

Hot-plate test

***

On On+NaloxoneOff

****

0

20

40

60

Day 1Off

Day 2On

Day 3 Off

Hot-plate testLate

ncy

to li

ck p

aws

(sec

)

0

20

40

60

ControlTg POMC-ChR2

********

Food

inta

ke (g

ram

/100

g B

W)

0.0

0.5

1.0

1.5

2.0

2.5

Day 1Off

Day 2On

Day 3 Off

******** ControlTg POMC-ChR2

A B C

Fig. 4. Activation of Arc neurons derived from POMC-expressing lineage increases food intake in Tg POMC-ChR2 mice. (A) Repeated-measures two-wayANOVA revealed a significant genotype × day interaction, F(2, 82) = 9.44, P < 0.001, for the hot-plate test. Blue light stimulation of Arc neurons in Tg POMC-ChR2 mice led to an increased latency for the animals to lick their paws, F(2, 52) = 21.91, P < 0.0001; ****P < 0.0001; n = 27 mice. Latency for single transgeniclittermate control mice without ChR2-eYFP expression to lick their paws was unchanged, F(2, 30) = 1.06, P = 0.36; n = 16 mice, 8 mice per single transgenicmouse line. (B) Increased latency induced by light stimulation in Tg POMC-ChR2 mice was reduced by pretreatment with opioid antagonist naloxone,F(2, 24) = 15.2, P < 0.0001; ***P < 0.001, ****P < 0.0001; n = 13 mice. (C) Repeated-measures two-way ANOVA revealed a significant genotype × day interaction,F(2, 54) = 76.59, P < 0.0001, for the food intake study. Light stimulation of Arc neurons evoked a robust increase in food intake in Tg POMC-ChR2 mice,F(2, 24) = 68.28, P < 0.0001; ****P < 0.0001; n = 13 mice. Food intake for control mice was unchanged following light stimulation of Arc, F(2, 30) = 1.19, P = 0.32;n = 16 mice. Box plots show median, mean (+), lower and upper quartiles (boxes), and minima and maxima (whiskers).

**** ****

Food

inta

ke (g

ram

/100

g B

W)

Day 1 Day 2 Day 3 Off On Off

0.0

0.5

1.0

1.5 ** **

Late

ncy

to li

ck p

aws

(sec

)

Hot-plate testDay 1

OffDay 2

OnDay 3

Off

0

20

40

60*

**

Late

ncy

to li

ck p

aws

(sec

)

Hot-plate test

OffDay 1

OnDay 2

On+NaloxoneDay 3 0

20

40

60A B C

Fig. 3. Selective activation of POMC neurons in the Arc suppresses food intake in fasted viral POMC-ChR2 mice. (A) Blue light stimulation of POMC neuronsinduced a reduction of food intake in 4-h fasted viral POMC-ChR2 mice, F(2, 40) = 18.99, P < 0.0001; ****P < 0.0001; n = 21 mice. (B) Light stimulation of POMCneurons led to an increased latency for the mice to lick their paws during the hot-plate test, F(2, 30) = 8.82, P < 0.01; **P < 0.01; n = 16 mice. (C) Increasedlatency induced by light stimulation was blunted by pretreatment with opioid antagonist naloxone (10 mg/kg) in the hot-plate test, F(2, 28) = 6.797, P < 0.01;*P < 0.05; **P < 0.01; n = 15 mice. Repeated-measures one-way ANOVA followed by Turkey’s test were used for all above statistics. Box plots show median,mean (+), lower and upper quartiles (boxes), and minima and maxima (whiskers).

Wei et al. PNAS | vol. 115 | no. 40 | E9493

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 28

, 202

1

induction 0.21 versus 0.10 relative to respective control) (SIAppendix, Tables S3 and S4). Equally important, the cFos re-sponse in most brain regions was completely abolished by pre-treatment with naloxone (Fig. 7B and SI Appendix, Table S4),including cerebral cortex, PFC, CPu, NAcc, septum, PVN, andlateral thalamus. These results suggest that the acute increase in

food intake following activation of the Arc in Tg POMC-ChR2 miceis likely mediated by activation of the opioid receptor-signaling sys-tem with direct and indirect impact throughout the brain.

C 1

2

3

Off On

4

5

6

Off On

A

B

Fig. 5. Anatomical analysis of Arc POMC neuronal projections and the extent ofactivation in viral POMC-ChR2 mice. (A) eYFP-expressing Arc POMC neurons andtheir connectivity shown as a max intensity z-projection image. RepresentativeCLARITY processed image shows prominent arcuate connectivity with the para-ventricular, dorsomedial, and lateral hypothalamus, as well as the zona incerta,periaqueductal gray, and lateral septum in the viral POMC-ChR2mouse brain slice(4-mm thickness). (Scale bar, 1500 μm.) (B) Magnified view shows a more detailedeYFP expression pattern in the Arc in the viral-injected hemisphere. (Scale bar,600 μm.) (C) Neuronal activation in the brain following light stimulation in theArc of viral POMC-ChR2 mice. ISH revealed cFos mRNA expression was increasedin limited brain regions in response to selective activation of Arc POMC neurons.Images were quantitatively analyzed with results shown in SI Appendix, Table S2.Representative images are shown ranging from Bregma 1.10 mm (indicated bythe no. 1) through Bregma −2.46 mm (indicated by the no. 6).

C 1

2

3

4

5

6

7

8

Off On Off On

A

B

Fig. 6. Anatomical analysis of the projections of Arc neurons derived fromPOMC-expressing progenitors and the extent of activation in Tg POMC-ChR2mice.(A) Labeled Arc neurons and their connectivity shown as a max intensity z-projection image. Representative image shows prominent arcuate connectivitywith the dorsomedial, paraventricular, and lateral hypothalamus in the Tg POMC-ChR2mouse brain slice (2-mm thickness) processed with the iDISCOmethod. (Scalebar, 1000 μm.) (B) Magnified view shows detailed eYFP expression pattern in theArc. (Scale bar, 500 μm.) (C) ISH revealed cFos mRNA expression was intensivelyincreased in broader brain areas in response to activation of Arc neurons of TgPOMC-ChR2 mice. Images were quantified with results shown in SI Appendix,Table S3. Representative images are shown ranging from Bregma 1.10 mm (in-dicated by the no. 1) through Bregma −2.12 mm (indicated by the no. 8).

E9494 | www.pnas.org/cgi/doi/10.1073/pnas.1802237115 Wei et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 28

, 202

1

DiscussionIn this study, we examined the relationship between cell-type–specific neuron activation and feeding behavior by optogeneticstimulation of both mature POMC neurons and neurons derivedfrom POMC-expressing progenitors in the Arc of the hypothal-amus. This body of work revealed the following main findings: (i)selective activation of POMC neurons rapidly inhibits feeding infood restricted animals; (ii) simultaneous activation of bothPOMC and orexigenic neurons derived from a common pro-genitor, including a subset of AgRP neurons, leads to a robustincrease in feeding; (iii) 3D imaging delineates the projectionpattern of adult hypothalamic POMC neurons versus the pro-jection pattern of neurons derived from embryonic POMC-expressing cells that differentiate into POMC and a subset of

the AgRP population; (iv) brain regions induced by activation ofPOMC alone and associated with reduced feeding include thecerebral cortex, lateral septum, PVN, dorsomedial hypothala-mus, and the Arc; (v) coupling of neuronal activation andpharmacological treatment shows that activation of orexigenicneurons derived from POMC-expressing progenitors engages anextensive neural network that involves the endogenous opioidsystem. Thus, the increased food intake was blunted by pre-treatment with naloxone, an opioid receptor antagonist. This wasaccompanied by the normalization of neural activity in most ofthe activated brain regions, including the cerebral cortex, PFC,CPu, NAcc, septum, PVN, and lateral thalamus.In the Arc, posttranslational processing of the POMC pro-

hormone results in two primary peptide products, α-MSH andβ-endorphin (11). β-Endorphin is involved in pain regulation(12) and α-MSH mediates the anorexigenic effect on feeding (7,25–29). In this study, both viral POMC-ChR2 and Tg POMC-ChR2 mice showed an analgesic effect induced by light stimu-lation, demonstrating functional activation of POMC neurons inthese two animal models. Selective stimulation of POMC neu-rons in viral POMC-ChR2 mice led to decreased food intake,consistent with the observation that the hypothalamic melano-cortin pathway is involved in feeding behavior (9, 10). This acuteinhibition of feeding in fasted viral POMC-ChR2 mice is alsoconsistent with pharmacological studies (25–29). Thus, an α-MSHagonist has a potent inhibitory effect on deprivation-induced foodintake within 1 or 2 h (28, 29) when administrated in the PVN,a site that receives projections from Arc POMC neurons andwhere MC4-R expression is very high. However, our finding is inpartial contrast with previous studies using optogenetic (30) orchemogenetic (31, 32) activation of POMC neurons. While thesestudies demonstrate reduced food consumption, they report thatthe effect is not evident until 24 h of stimulation in ad libitum-fedmice (30–32). One possibility for this discrepancy is that we useda different light-stimulation pattern in our study. Another dif-ference is that our mice were fasted for 4 h, which may con-tribute to the more immediate effects of stimulation on feedingbehavior. Our results suggest that Arc POMC neurons exertfaster regulation of satiety in food-motivated mice. This findingis supported by a recent report that at least some Arc POMCneurons may act on faster time scales to reduce the drive to seekfood (33). The combination of these pharmacological and opto-or chemogenetic studies would lead to a more complex view ofthe role of POMC in food regulation, whereby it would exert along-term modulatory role based on metabolic status in thefreely feeding animals, but is also active on shorter time scalesduring times when the animal is motivated to eat and the mon-itoring of food intake by these appetitive circuits is more critical.This does not preclude other mechanisms that recruit POMC toinduce eating cessation (31, 32).Surprisingly, in Tg POMC-ChR2 mice, activation of Arc

neurons derived from POMC-expressing progenitors evokedvoracious feeding. We have demonstrated that in these animals,ChR2 is expressed in both POMC- and AgRP-positive neurons,and these two groups of neurons are simultaneously activated asdetermined by cFos protein induction. The POMC neurons inthe Arc were also functionally activated as evidenced by the painregulation experiment in this animal model. The increased foodintake in Tg POMC-ChR2 mice demonstrates that activation ofthis subset group of orexigenic neurons derived from POMC-expressing progenitors overrides the anorexigenic effect con-tributed by POMC neuron activation. It is well established thatPOMC and AgRP/NPY neurons in the Arc exert opposing ac-tions on feeding to maintain energy homeostasis in the body (8,14). Genetic and pharmacological evidence suggests that AgRP/NPY neurons promote feeding and increase body weight (8–10).Opto- or chemogenetic stimulation of AgRP neurons promotesintense food consumption (30, 34–36). The opposing effects on

B 1

2

3

4

5

6

7

8

Off On+Naloxone Off On+Naloxone

A

0.0

0.5

1.0

1.5

2.0

Off On On+Naloxone(2 mg/kg)

On+NaloxoneOff On(10 mg/kg)

Food

inta

ke (g

ram

/100

g B

W)

**

Fig. 7. Involvement of endogenous opioids in the increased feeding of TgPOMC-ChR2 mice. (A) Increased food intake in Tg POMC-ChR2 was reducedby pretreatment with naloxone in a dose-dependent manner (n = 6 mice in2 mg/kg naloxone group; n = 8 mice in 10 mg/kg naloxone group). A mul-tilevel regression model was used to assess the effects of light stimulation,naloxone treatment, and naloxone dosage on food intake, as well as aconditional effect of dosage on naloxone treatment (naloxone × dosage).Light stimulation increased food intake, β = 1.08, T(25) = 12.26, P < 0.0001, ina manner that was opposed by naloxone treatment at 2 mg/kg, β = −0.33,T(25) = −2.61, *P < 0.05. The ability of naloxone to oppose the effects ofthe light stimulation was even stronger in the group of mice treated with ahigher dosage of naloxone, β = −0.42, T(25) = −2.71, *P < 0.05; 10 mg/kgversus 2 mg/kg. Otherwise food intake in these two groups of mice wassimilar, β = 0.04, T(12) = 0.27, P = 0.79. Mean ± SEM, 2 mg/kg naloxonegroup: 0.13 ± 0.02 (light off, saline); 1.13 ± 0.23 (light on, saline); 0.84 ± 0.18(light on, naloxone); 10 mg/kg naloxone group: 0.092 ± 0.024 (light off,saline); 1.24 ± 0.11 (light on, saline); 0.46 ± 0.069 (light on, naloxone). (B) ISHrevealed that cFos mRNA expression was either normalized or reduced inbrain regions involved in reward and motivation by pretreatment withopioid antagonist naloxone (10 mg/kg). Images were quantified with resultsshown in SI Appendix, Table S4. Representative images are shown rangingfrom Bregma 1.10 mm (indicated by the no. 1) through Bregma −2.12 mm(indicated by the no. 8).

Wei et al. PNAS | vol. 115 | no. 40 | E9495

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 28

, 202

1

food consumption following optogenetic activation in thetransgenic and the viral models and their differences in potencystand in sharp contrast to the effects on analgesia. Thus, opto-genetic stimulation in both the viral and the transgenic modelsproduced comparable analgesic potency. This is likely because,in both cases, the analgesia is elicited from the activation ofPOMC neurons only, whereas the food regulation is the result ofthe relative power of POMC versus AgRP/NPY neurons. Takentogether, our finding of increased food intake in Tg POMC-ChR2 mice suggests that mechanisms that increase satiety maynot be as effective under conditions that enhance the inhibitionof food consumption and oppose AgRP.The increased food intake in Tg POMC-ChR2 mice can be

largely blocked by pretreatment with the nonselective opioidreceptor antagonist naloxone, suggesting involvement of theopioid receptor-signaling system. Other neural systems may alsocontribute to this increased food intake because naloxone didnot completely block the increased food intake. Opioids are afamily of peptides that differentially interact with three receptorclasses (37). Anatomical studies have demonstrated that opioidreceptors and peptides are widely distributed throughout thebrain (38), including in a number of hypothalamic nuclei. Giventhe wide distribution of opioid receptors and the presence ofother neuropeptides in the vicinity of opioidergic pathways,opioids are in position to affect many aspects of feeding andenergy homeostasis (39, 40). Pharmacological studies have in-dicated that opioids stimulate food intake, whereas opioid an-tagonists are inhibitory (39, 40). It should be noted that thisincreased food intake in Tg POMC-ChR2 mice may not be solelydue to the release of endogenous β-endorphin. Selective acti-vation of POMC neurons in unsated viral POMC-ChR2 miceinduced a reduction of food intake despite the possibility thatendogenous β-endorphin was likely released as evidenced by ananalgesic effect during the hot-plate test.We then investigated the possible neural mechanisms whereby

activation of Arc neurons derived from POMC-expressing pro-genitors led to increased food intake while selective activation ofPOMC neurons resulted in decreased food intake using cFosmRNA as a marker. Selective activation of POMC neurons inviral POMC-ChR2 mice led to neural activation in limited brainregions that are typically thought of as regulating food intake(41), including cerebral cortex, lateral septum, PVN, dorsome-dial hypothalamus, and Arc. Classic lesion studies in PVN,dorsomedial, or ventromedial hypothalamus induced hyperpha-gia and obesity, while electrical stimulation increased food intake(6). In addition, we found increased neural activity in cerebralcortex and lateral septum following stimulation of Arc POMCneurons, indicating their possible involvement in regulatingfood intake.In contrast, activation of the Arc in Tg POMC-ChR2 mice

resulted in an intensive cFos response in broader areas of thebrain thought to modulate feeding behavior (8, 42–44), includingcerebral cortex, PFC, NAcc, septum, PVN, LH, and dorsomedialhypothalamus. More importantly, the neuronal activity in cere-bral cortex, PFC, CPu, NAcc, septum, and PVN was normalizedby pretreatment with naloxone, while the neuronal activity of theLH was reduced by 52% with the same pretreatment. Several ofthese responsive regions, such as the PFC, the NAcc, and theLH, have connections with reward systems and motivated be-havior that modulate feeding (42–44). For example, activation ofdopamine D1 receptor-expressing neurons in the PFC increasesfood intake (45). The NAcc functions as a major projection areafor food reward and motivation in the mesolimbic dopaminergicsystem (42). The LH is known to be a critical neuroanatomicalsubstrate for motivated feeding behavior in corticostriatal–hypothalamic circuitry (43). Thus, the activation of the subset oforexigenic neurons derived from POMC-expressing progenitors

in Tg POMC-ChR2 mice may influence the rewarding charac-teristics of food and/or the motivational component of feeding.We also investigated the architecture of neural circuits spe-

cifically originating from Arc peptidergic neurons on a brain-wide scale in these two animal models via 3D visualization. In theviral POMC-ChR2 mice, CLARITY processing showed that ArcPOMC neurons project to a number of brain regions (SI Ap-pendix, Table S1), consistent with previous anatomical studies(46, 47). Our detailed mapping of Arc POMC neuronal con-nectivity points to functional studies on the roles and circuitmechanisms underlying the regulation of feeding in these areas.The apparent diffuse projections of Arc POMC terminals ob-served suggest that POMC participates in the regulation of foodconsumption not only through hypothalamic structures, but alsoextrahypothalamic nuclei, including the amygdala, zona incerta,and lateral septum. More specifically, the amygdala is of interestas it is an area implicated in emotional aspects of feeding be-havior. Recent studies have also shown that activation of zonaincerta GABAergic neurons results in rapid binge-like eatingbehavior (48), while inactivation of lateral septum neuronsblunts ventral hippocampus-induced suppression of feeding (49).However, it is still not fully understood how the inputs from ArcPOMC neurons affect the neuronal physiology in these brainregions or their potential regulation of feeding behavior. Incontrast to the viral POMC-ChR2 mice, a seemingly greaterintensity of hypothalamic eYFP signal in the Tg POMC-ChR2mice necessitated diminishing the comparatively less intenseextrahypothalamic projections through the image thresholdfunction. Nevertheless, the 3D visualization of neuronal projec-tions in Tg POMC-ChR2 mice highlights the extensive neuronalcommunication between the Arc and lateral hypothalamic nu-cleus which is at the center of feeding control.In sum, the POMC-expressing progenitors in the Arc differ-

entiate into two functionally distinct neuronal subpopulations toinfluence feeding behavior in adulthood. Furthermore, the dif-ferences in food intake are associated with distinct projectionand activation patterns of Arc neurons derived from POMC-expressing lineages compared with adult hypothalamic POMCneurons. Translational applications that aim to control appetitelikely need to target the activation rather than the inhibitionmechanisms and to modulate either the AgRP system itself or itsdownstream mediators. The fact that increased food intake in TgPOMC-ChR2 mice can be blunted by the opioid antagonistnaloxone coupled with the normalization of neuronal activity inbrain regions involved in reward and motivation highlights thepotential value of targeting opioid receptor signaling to controlfeeding behavior.

Materials and MethodsAnimals. POMC-Cre and Rosa26-CAG-stopflox-ChR2(H134R)-eYFP (Ai32) micewere generated as previously described (16, 17). All experiments were con-ducted in accordance with the principles and procedures outlined in theNational Institutes of Health Guidelines for the Care and Use of Animals andwere approved by the Institutional Animal Care and Use Committee at theUniversity of Michigan. More detailed information is provided in SI Ap-pendix, SI Material and Methods.

Stereotactic Surgery and Virus Vector Injection. For the Tg POMC-ChR2 mice,optical fiber was unilaterally placed above the right Arc [coordinates,Bregma: (anterior–posterior, AP) −1.62 mm, (dorso–ventral, DV) 5.60 mm,(medio– lateral, ML) 0.33 mm]. For the POMC-Cre mice, Cre-dependentdouble-floxed deno-associated virus AAV5-EF1α-DIO-ChR2(H134R)-eYFP (50)was injected unilaterally in the right Arc [coordinates, Bregma: (AP)−1.62 mm, (DV) 5.70 mm, (ML) 0.30 mm]. Following injection, optical fiberwas implanted over the Arc [(AP) −1.62 mm, (DV) 5.45 mm, (ML) 0.30 mm].This mouse line of POMC-Cre with selectively expression of ChR2 in adult-hood by viral-mediated method is referred to as viral POMC-ChR2. For de-tails see SI Appendix, SI Material and Methods.

E9496 | www.pnas.org/cgi/doi/10.1073/pnas.1802237115 Wei et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 28

, 202

1

Slice Electrophysiology. For whole-cell current-clamp recordings, ChR2-eYFP-positive neurons in the Arc from acute slices were stimulated with repeatedlight-stimulation patterns used in the hot-plate and food intake tests of thecurrent study. For details see SI Appendix, SI Material and Methods.

Immunohistochemistry. Staining of brain sections from viral POMC-ChR2 andTg POMC-ChR2 mice with anti-GFP, anti-POMC, anti-AgRP, anti–β-endorphin,and anti-cFos was done as described in SI Appendix, SI Material and Meth-ods. For all cFos protein induction experiments, mice received 6 min of bluelaser (473 nm) stimulation (10 Hz, 15-ms pulses, 4.3 mW on the tip of fiber) inthe open field chamber, then returned to their home cage. Animals wereperfused 90 min after the onset of light stimulation. For details see SI Ap-pendix, SI Material and Methods.

Behavioral Experiments. All behavioral tests were conducted at least 3-wkpostsurgery during the same circadian period (0900–1600). Mice were han-dled and attached to the optical fiber without photostimulation for 4 dbefore behavioral tests. For the hot-plate test, the latency (in seconds) forthe mice to lick their paws was measured (cutoff time, 60 s). On 3 consecu-tive testing days, mice were connected to the optical fiber and placed in anempty mouse cage for 5 min (day 1, blue laser off; day 2, blue laser on; day 3,blue laser off). Immediately following this 5-min period, mice were thentested on the hot plate. Blue laser stimulation (15-ms pulses, 10 Hz, 4.3 mWon the tip of fiber) on day 2 continued until the completion of the test. Fornaloxone pretreatment studies, mice were injected intraperitoneally withsaline or naloxone HCl (10 mg/kg) 15 min before laser stimulation.

For the regular food intake test, mice had ad libitum access to standardlaboratory pellet food in the home cage before and after tests. Each mousewas tested at the same time each day. Food consumption was measured for30 min for 3 consecutive days (day 1, laser off; day 2, blue laser on; day 3, bluelaser off). The laser stimulation parameters used on day 2 were 15-ms lightpulses at 10 Hz for 30 s every minute for 30 min with 4.3 mW light power onthe tip of fiber. For naloxone pretreatment in the Tg POMC-ChR2 mouse line,animals were injected intraperitoneally with saline or naloxone (2 mg/kg;10 mg/kg) 15 min before laser stimulation. For the food intake test in viralPOMC-ChR2 mice, animals were food deprived for 4 h with water freelyavailable before the regular food intake test each day. The laser stimulationon day 2 was the same as described above. More detailed information isprovided in SI Appendix, SI Material and Methods.

Three-Dimensional Visualization. For 3D imaging of immunofluorescence inviral POMC-ChR2 mice, the CLARITY procedure (19–21) was followed (SIAppendix, SI Material and Methods). For 3D imaging of immunofluores-cence in Tg POMC-ChR2 mice, the iDISCO method (22) was followed (SI

Appendix, SI Material and Methods). CLARITY optimized light sheet micro-scope (COLM) was used to acquire the images (20). Tiles of acquired imagestacks were stitched together using TeraStitcher and subsequently visualizedin 2D through orthoslice mode and in 3D through maximum intensity pro-jection volume rendering mode using Amira 3D software.

cFos mRNA Expression Following Optogenetic Stimulation. For all cFos mRNAresponse experiments, mice received 6 min of either blue laser stimulation(10 Hz, 15-ms pulses, 4.3 mW on the tip of fiber) or no laser stimulation in theopen field chamber, and then were returned to their home cage. Mice werekilled by rapid decapitation 30 min after the onset of laser stimulation andtheir brains were dissected. For naloxone pretreatment in the Tg POMC-ChR2 mouse line, animals were injected intraperitoneally with naloxone(10 mg/kg) 15 min before laser stimulation. In situ hybridization (ISH)methodology was performed as previously described (51). For details see SIAppendix, SI Material and Methods.

Statistical Analysis. Statistical analyses were performed using Prism 7.0(GraphPad) software or R software. P < 0.05 was considered statisticallysignificant. Experimental animals demonstrating inaccurate viral injection orcannula sites were excluded from the present study since these subjectslacked expression of the reporter gene or neuronal activity marker cFos.These validation parameters were established before data analysis. For thenaloxone pretreatment study on feeding behavior in Tg POMC-ChR2 mice,a multilevel regression model (also known as a hierarchical linear model ora mixed-effects model) was used to assess the effects of light, naloxonetreatment, and naloxone dosage on food intake, while accounting forwithin-subjects design. These analyses were performed in R (R v.3.3.0: https://cran.r-project.org; RStudio v.0.99.896: www.rstudio.com/) using the lme()function in the nlme package (https://CRAN.R-project.org/package=nlme)with a default correlation structure (no within-group correlation) andmethod (restricted maximum likelihood). Our model included the fixed ef-fects of the factors light (reference level = off), naloxone treatment (refer-ence level = no naloxone), and dosage (reference level = 2 mg/kg), as well asa conditional effect of dosage on naloxone treatment (naloxone × dosage),with repeated measures accounted for by including individual subjects as arandom effects variable (random intercept).

ACKNOWLEDGMENTS. We thank Drs. Megan Hagenauer and Aram Parsegianfor help on statistical analysis; Dr. Cortney Turner for helpful discussion ofthe manuscript; and Limei Zhang and Hui Li for technical assistance. Thiswork was supported by NIH Grants R01MH104261 and ONR N00014-12-1-0366, the Hope for Depression Research Foundation, and the Pritzker Neu-ropsychiatric Research Consortium.

1. Ogden CL, et al. (2006) Prevalence of overweight and obesity in the United States,1999–2004. JAMA 295:1549–1555.

2. Olszewski PK, Levine AS (2007) Central opioids and consumption of sweet tastants:When reward outweighs homeostasis. Physiol Behav 91:506–512.

3. Sominsky L, Spencer SJ (2014) Eating behavior and stress: A pathway to obesity. FrontPsychol 5:434.

4. Flier JS, Maratos-Flier E (1998) Obesity and the hypothalamus: Novel peptides for newpathways. Cell 92:437–440.

5. Elmquist JK, Elias CF, Saper CB (1999) From lesions to leptin: Hypothalamic control offood intake and body weight. Neuron 22:221–232.

6. Hoebel BG, Teitelbaum P (1962) Hypothalamic control of feeding and self-stimulation.Science 135:375–377.

7. Cone RD (2005) Anatomy and regulation of the central melanocortin system. Nat

Neurosci 8:571–578.8. Morton GJ, Meek TH, Schwartz MW (2014) Neurobiology of food intake in health and

disease. Nat Rev Neurosci 15:367–378.9. Ollmann MM, et al. (1997) Antagonism of central melanocortin receptors in vitro and

in vivo by agouti-related protein. Science 278:135–138.10. Cowley MA, et al. (1999) Integration of NPY, AGRP, and melanocortin signals in the

hypothalamic paraventricular nucleus: Evidence of a cellular basis for the adipostat.

Neuron 24:155–163.11. Akil H, Watson SJ (1980) Neuromodulatory functions of the brain pro-opiocortin

system. Adv Biochem Psychopharmacol 22:435–445.12. Loh HH, Tseng LF, Wei E, Li CH (1976) beta-endorphin is a potent analgesic agent.

Proc Natl Acad Sci USA 73:2895–2898.13. Millington GWM (2007) The role of proopiomelanocortin (POMC) neurones in feed-

ing behaviour. Nutr Metab (Lond) 4:18.14. Schwartz MW, Porte D, Jr (2005) Diabetes, obesity, and the brain. Science 307:

375–379.15. Padilla SL, Carmody JS, Zeltser LM (2010) Pomc-expressing progenitors give rise to

antagonistic neuronal populations in hypothalamic feeding circuits. Nat Med 16:403–405.

16. McHugh TJ, et al. (2007) Dentate gyrus NMDA receptors mediate rapid pattern sep-aration in the hippocampal network. Science 317:94–99.

17. Madisen L, et al. (2012) A toolbox of Cre-dependent optogenetic transgenic mice forlight-induced activation and silencing. Nat Neurosci 15:793–802.

18. Akil H, Young E, Walker JM, Watson SJ (1986) The many possible roles of opioids andrelated peptides in stress-induced analgesia. Ann N Y Acad Sci 467:140–153.

19. Chung K, Deisseroth K (2013) CLARITY for mapping the nervous system. Nat Methods10:508–513.

20. Tomer R, Ye L, Hsueh B, Deisseroth K (2014) Advanced CLARITY for rapid and high-resolution imaging of intact tissues. Nat Protoc 9:1682–1697.

21. Yang B, et al. (2014) Single-cell phenotyping within transparent intact tissue throughwhole-body clearing. Cell 158:945–958.

22. Renier N, et al. (2014) iDISCO: A simple, rapid method to immunolabel large tissuesamples for volume imaging. Cell 159:896–910.

23. Anderson EJ, et al. (2016) 60 YEARS OF POMC: Regulation of feeding and energyhomeostasis by α-MSH. J Mol Endocrinol 56:T157–T174.

24. Overstreet LS, et al. (2004) A transgenic marker for newly born granule cells indentate gyrus. J Neurosci 24:3251–3259.

25. Fan W, Boston BA, Kesterson RA, Hruby VJ, Cone RD (1997) Role of melanocortinergicneurons in feeding and the agouti obesity syndrome. Nature 385:165–168.

26. Rossi M, et al. (1998) A C-terminal fragment of Agouti-related protein increasesfeeding and antagonizes the effect of alpha-melanocyte stimulating hormone invivo. Endocrinology 139:4428–4431.

27. Nakazato M, et al. (2001) A role for ghrelin in the central regulation of feeding.Nature 409:194–198.

28. Giraudo SQ, Billington CJ, Levine AS (1998) Feeding effects of hypothalamic injectionof melanocortin 4 receptor ligands. Brain Res 809:302–306.

29. Cowley MA, et al. (2001) Leptin activates anorexigenic POMC neurons through aneural network in the arcuate nucleus. Nature 411:480–484.

30. Aponte Y, Atasoy D, Sternson SM (2011) AGRP neurons are sufficient to orchestratefeeding behavior rapidly and without training. Nat Neurosci 14:351–355.

31. Zhan C, et al. (2013) Acute and long-term suppression of feeding behavior by POMCneurons in the brainstem and hypothalamus, respectively. J Neurosci 33:3624–3632.

Wei et al. PNAS | vol. 115 | no. 40 | E9497

NEU

ROSC

IENCE

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 28

, 202

1

32. Fenselau H, et al. (2017) A rapidly acting glutamatergic ARC→PVH satiety circuitpostsynaptically regulated by α-MSH. Nat Neurosci 20:42–51.

33. Mandelblat-Cerf Y, et al. (2015) Arcuate hypothalamic AgRP and putative POMCneurons show opposite changes in spiking across multiple timescales. eLife 4:07122.

34. Krashes MJ, et al. (2011) Rapid, reversible activation of AgRP neurons drives feedingbehavior in mice. J Clin Invest 121:1424–1428.

35. Atasoy D, Betley JN, Su HH, Sternson SM (2012) Deconstruction of a neural circuit forhunger. Nature 488:172–177.

36. Chen Y, Lin YC, Zimmerman CA, Essner RA, Knight ZA (2016) Hunger neurons drivefeeding through a sustained, positive reinforcement signal. eLife 5:18640.

37. Akil H, et al. (1984) Endogenous opioids: Biology and function. Annu Rev Neurosci 7:223–255.

38. Mansour A, Fox CA, Akil H, Watson SJ (1995) Opioid-receptor mRNA expression in therat CNS: Anatomical and functional implications. Trends Neurosci 18:22–29.

39. Kalra SP, Horvath TL (1998) Neuroendocrine interactions between galanin, opioids,and neuropeptide Y in the control of reproduction and appetite. Ann N Y Acad Sci863:236–240.

40. Glass MJ, Billington CJ, Levine AS (1999) Opioids and food intake: Distributed func-tional neural pathways? Neuropeptides 33:360–368.

41. Gao Q, Horvath TL (2007) Neurobiology of feeding and energy expenditure. AnnuRev Neurosci 30:367–398.

42. Kelley AE, Baldo BA, Pratt WE, Will MJ (2005) Corticostriatal-hypothalamic circuitryand food motivation: Integration of energy, action and reward. Physiol Behav 86:773–795.

43. Stuber GD, Wise RA (2016) Lateral hypothalamic circuits for feeding and reward. NatNeurosci 19:198–205.

44. Saper CB, Chou TC, Elmquist JK (2002) The need to feed: Homeostatic and hedoniccontrol of eating. Neuron 36:199–211.

45. Land BB, et al. (2014) Medial prefrontal D1 dopamine neurons control food intake.Nat Neurosci 17:248–253.

46. Bagnol D, et al. (1999) Anatomy of an endogenous antagonist: Relationship betweenAgouti-related protein and proopiomelanocortin in brain. J Neurosci 19:RC26.

47. Wang D, et al. (2015) Whole-brain mapping of the direct inputs and axonal projec-tions of POMC and AgRP neurons. Front Neuroanat 9:40.

48. Zhang X, van den Pol AN (2017) Rapid binge-like eating and body weight gain drivenby zona incerta GABA neuron activation. Science 356:853–859.

49. Sweeney P, Yang Y (2015) An excitatory ventral hippocampus to lateral septum circuitthat suppresses feeding. Nat Commun 6:10188.

50. Tsai HC, et al. (2009) Phasic firing in dopaminergic neurons is sufficient for behavioralconditioning. Science 324:1080–1084.

51. Aurbach EL, et al. (2015) Fibroblast growth factor 9 is a novel modulator of negativeaffect. Proc Natl Acad Sci USA 112:11953–11958.

E9498 | www.pnas.org/cgi/doi/10.1073/pnas.1802237115 Wei et al.

Dow

nloa

ded

by g

uest

on

Feb

ruar

y 28

, 202

1